TW583560B - Modeling and scheduling of a semiconductor wafer fab - Google Patents

Abstract

In this patent, we propose a simulation/scheduling system for semiconductor wafer fab. The underlying model is basically constructed with colored timed Petri-net. The resource contention subnet in this model is modified with the help of the queueing theory. We apply the queueing theory to simplify the simulation process and to reduce simulation time. Besides, an arrival-rate-updating module is designed to reflect the dynamic flows of lots in the fab. This module can be used to improve the accuracy of simulation. Furthermore, real system behaviors such as machine breakdown, machine maintenance and lot transportation are involved in our detailed model. Our simulator is built on this model. The genetic algorithm is adopted in our scheduler for its well-known efficient searching ability in a huge solution space. The evaluation process in the genetic algorithm is completed by our simulator. We also propose a gene encoding method which can provide versatile scheduling strategies and support traditional rules.

Description

ILTJ Outfit -------- Order --------- Line < (Please read the precautions on the back before filling out this page) Printed by the Intellectual Property Bureau Employee Consumer Cooperative of the Ministry of Economic Affairs 583560 A7- -——--------- B7_______ V. Description of the invention (1) Background of the invention In the early stage of the proposed production, a good model construction is undoubtedly necessary. Following a complete and detailed model can guarantee the accuracy of the simulator. The Petri net in learning # has been recognized as a weight-based tool for describing and analyzing complex discrete event systems. Analysis on the Fischer network usually incorporates simulations, but simulations often take a significant amount of time. On the other hand, some designs are based on analytical techniques, but they are often too complicated for practical use. A service system composed of queues depicts more interesting dynamic systems. Tsai Lie systems are often used as tools for modelling and performance analysis of communication systems, mainly because they are usually more efficient than other methods. Both systems have their own advantages and limitations. For example, although the Fischer T-Road cannot easily reflect the situation of the side system, it can easily describe the behavior of synchronization and / or decentralization and reorganization (work). In contrast, it is difficult to describe this using pure queue system. Kind of situation. If we can combine the description of discrete events and the mathematical analysis of the queue system with the Fischer network, it means that we can build a model that approximates the real system on the one hand, and does not Precision at the expense of efficiency. This patent simulator is designed with this in mind. & In past research and design, enqueuing systems often assumed that their conditions were under steady state. In the actual manufacturing plant, the flow of the batch may be unstable and change. Therefore, the above-mentioned simulator needs a mechanism that can reflect the rate at which the batch of goods arrive at the machine. In addition to the true performance of the manufacturing i behavior, it can also improve the accuracy of the queue system production behavior. This genetic algorithm (genetic algorithm) is a paper scale commonly used to solve empty papers _China National Standard (CNSU4 Specification ⑵Q χ Norwegian public issued) 583560 Intellectual Property Bureau of the Ministry of Economic Affairs Employee Cooperative Cooperative printed A7, Invention Description (2) Algorithms for interim best solution search. Research and application of manufacturing charm production schedule. However, the application of genetic algorithms has two problems: Ai is the solution to evaluate actions, and the second is gene coding. Both of these affect the efficiency and quality of the genetic algorithm. The action of solution evaluation usually depends on simulation. The biggest disadvantage of past simulation systems is that it is time consuming, so that the genetic gene has not searched for sufficient solutions in a limited time. As for the way of gene coding, it is more common to use the rule of preference. However, after combining the concepts of the queue system, we need a coding method that is compatible with the queue system. ~ Summary of Invention Semiconductor manufacturing is a high-risk industry. In order to survive under fierce competition, in addition to the need to accurately control the status of the factory at any time, it is also hoped that it can evaluate the efficiency of the manufacturing factory or conduct batch shipments within the factory. Delivery forecast and use it to win customers. Ding Dan, or change the factory settings. What's more, in the face of complex and lengthy processes, random shutdowns in manufacturing plants, high-priced equipment and product diversity, artificial production schedules are no longer sufficient. The industry is also looking forward to increasing the productivity of the manufacturing plant and the delivery rate of the products through automated scheduling. The patent system was developed in response to such needs. The present invention has three features. The first is to closely combine the colored timed Petri-net and queuing theory, which effectively reduces the time complexity of the simulation. The second is a detailed underlying model of the Fibonacci network, which simulates most of the production behavior of the wafer fab. The third is to propose a gene coding method for genetic algorithms, which can not only generate a variety of scheduling strategies, but also support the existing scheduling rules commonly used in manufacturing plants. This paper size applies to China National Standard (CNS) A4 (210 X 297 mm) J Pack -------- Order --------- Line · (Please read the note on the back first? Matters Fill out this page again) 583560 A7 B7 V. Description of the invention (3) Detailed description of the invention ♦ The underlying model ~ Before the development of the proposal ||, the establishment of the model is quite critical. The good intentions of the model are related to the efficiency and accuracy of the simulation. The underlying model of the patented estimator is established by the Fischer network combined with queuing theory. The Fibonacci network is a well-known graphical tool for constructing peppers, and the queuing theory is a set of mathematical principles for analyzing the row system. The following will describe the actual operation behavior of the plant in which the patent includes the model towel, that is, the model Accuracy; and how to simplify the model with queuing theory to increase pseudo-efficiency. The QCPN model extension of the sequel-colored plus-time plus Fischer-Hertz network (QCPN) model proposed here introduces the symbol 歹 _ ^ in the specific place. This part will be explained in the section of queuing theory. In the j Dinglie-style force port and color force port, the Fischer-Hertz network (QCpN) model is to simulate the underlying model of 11. The bottom model contains two main types: the reliance type and the transportation system model. The process model = the process model. To simulate production-related secret models and machine basic models in Simulation 4. Knife for clothing 1 · Process routing model The machine in the process of scrapping may have the above. Keer and Shou 'must first select a process capability machine according to the process steps. This paper ruler _ 家 标 ^ 297 mm) ( Please read the note on the back? Matters before filling out this page) -------- t --------- 583560 5. Description of the invention (4 is used to simulate this behavior The basic model of the machine The basic model of the machine includes _machine model test model and machine inspection_type = $, batch inspection ί !, V: r batch inspection model, etc. may occur during the operation of the platform. In some machines, after the batch of goods is processed, multiple inspection procedures are needed. | Mang Ma Guang A ~ Wang Duan's exhaustion scheduler, ± 仃 is. And machine inspection _ type is listed in the table. Events of scheduled maintenance • The model of the transportation system now shows that the number of process steps in the manufacturing process is huge. The number of batches must flow on so many machines, and the number is also completed. In the model of this patent, R Ran also considers the role of the shipping wheel system. The Wei model of the shipping wheel proposed here is in the making of obstacles. Routing, loading, and Wu Chen. The following briefly describes the behaviors modeled by each sub-model. First, private movement 1. Vehicle calling When a batch of goods needs to be moved from the current machine to a remote machine, the transportation demand is reduced, Request-vehicle to the service ^ transfer ~ transport 2. carrier selection This paper size turning + family standard (CNS) A4 specifications ⑵2977 mm) 583560 A7 5. Invention description (5) S transport system accepts one transport Electricity required-vehicle to go to service. The model of this patent ^; chooses the __ rule according to the vehicle selection strategy, trying to balance every-=!: ;; == is the secret ad 3 · vehicle service itinerary Each vehicle may also need to serve a stroke model to model Each service requires ^^ shop and requires the service of the spear force ^ ^ order. The simulation ⑼ uses 4 · Vehicle routing. The route from w to the destination machine may be sh—a system. Viewer use 5. Vehicle route conflicts Idle ’If there is—an idle vehicle possesses its route,; ==: the route. Vehicle Route Conflict Model-Model this. 6. Vehicle movement behavior. The moving die of the vehicle shows the charm of the 4 non-team theory that there will be acceleration and deceleration in the crane process System hiq) ^ Η, A7 V. Description of the invention (6) After the production is over, when the work is completed-Qi, if the machine is waiting for you, then the machine will do it. In the manufacturing plant, there will usually be order-bought goods & select Dubbe ’s priority rule, in order to calculate the processing priority of each arrival, each time the machine faces the selection, The next batch to be processed can be selected according to this priority. However, due to the arrival of new materials and materials, and the Japanese-Japanese fast-moving nature of the priority of shipment, the traditional model must consume the order of f and f. The queuing theory is attractive for solving this purpose. This book; = ^. In the past, such systems where the priority of shipments will change over time have been studied in depth. In addition-in the model proposed by this patent, there are several counties for the shipments. A priority level is provided with a pair of parameters and c ,. For each batch that belongs to the priority level /, its priority value in the column is calculated by bi + Cjx waiting time. The face of which is · g ίζ · Teng refers to the waiting time in the Zipai line. This excellent, equal_definition method can provide the rule of priority calculation for the loser, and it is also compatible with many previous common rules, such as FIFO, Last-in, last-out, least remaining process priority, shortest delivery time priority, etc. The first-in-first-out rule is listed below as an example. 4 (a) indicates the priority level of a certain type of batch, from an appropriate large value, // for all Number of priority levels, ^ is Μ Μ hci = z upon arrival of t〇ken, V / = 1, ..., // This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 χ 297 mm J— · --------------- ^- ------- ^ (Please read the Zhuyin on the back? Matters before filling out this page)

w P V: P, W,

b —-Mody, = 丨 A -—--- r _ 1- 'ί V CP ~~ Ci Γκ ~-Σ Two A c ~~ c Ί ~ ^ LL bt = /) + 1 1 Bureau of Intellectual Property, Ministry of Economic Affairs Printed by Employee Consumer Cooperative

The remaining processing time of the platform is a priority formula for the batch. For the batch of the parent in the queue, we «system efficiency. 2 ^^ === _ degree, the modification module is modified to the theory according to this article == Yes: _ contending network is making iso-sloping _ calculation formula, the rejection rate of batch t is — number (consistent with A Proportional), in order to reflect the second precision within the plant immediately, this patent has designed a batch · reach rate yf4. The effect of waiting time is to accumulate the number of arrivals of the batch, and regularly update the batch arrival :: rate: group simulator This paper size is applicable to Zhongguanjia Standard (CNS) A4 Rule Shijie x 297 Public Love '

I. _Binding --------- Line j (Please read the notes on the back before filling this page) 583560 A7 V. Invention Description (8) After establishing the underlying model, I need a set of tools to Integrate the model and the manoeuvres in it, and this is exactly what the simulation engine does. The shell materials in the simulation engine are mainly divided into two categories, one is the physical attributes of the item, such as machine position, machine process capability, machine process time, etc., and the other is control rules, such as machine selection rule, priority Level parameters, vehicle related laws. This information can be obtained by E-manufacturing executives, or set by the factory manager. It is worth mentioning that due to the flexibility and modularity of the underlying model, it is not necessary to modify the engine configuration for different manufacturers. The actual situation of the manufacturer can be simply reflected by the part of the material stream. The application scope of the simulator of this patent is greatly improved. The meticulousness of the underlying model and the integration of the queuing theory also make the accuracy and efficiency of the simulator meet the needs of the industry, which can respond to the production status of the entire plant in a short period of time, and provide the basis for each strategy. In addition to the degree of deduction, the model's meticulousness brings another characteristic. Operators can understand the details of Zhongmu in detail: for __simulation_take_factory__approval or __machine-off-state. Another important feature of the simulator is that it can predict the batch period. The model of this patent accepts the input of the manufacturer—a policy for the receipt and rejection of a batch of goods. If the simulation is based on Wei ’s strategy and the current plant and slate conditions, it is more important to predict the future period of time. The payment of the delivery of batches is not important to the manufacturing industry. K is not only an assessment of the manufacturing capacity, but also an important basis for contending for orders. ♦ Scheduler In a manufacturing plant, the most important indicators are the total production time of batches, manufacturing capacity, and the delivery rate of batches. In addition to purchasing more diversified and more efficient equipment X 297 mm) Please read t 1 1 丨 read the back 1 of the | I note I do not know the matter 1 | 顼 扉 save Φ Fill out the book • Page 1 | 1 1 1 1 1 Book size _ 家 鲜 （cns) A4 ^^ 583560

Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs

The most critical factor is the suspicion of the simple soil column. Generally speaking, the batch production schedule includes a batch purchase strategy and machine selection strategy. In this _convenience design—scheduled _ = combination of strategy and approval priority rule, and then satisfy the manager ’s ideal ir value schedule search and schedule evaluation, the scheduler of this patent ^ uses Xiao Chuan The basic materials method is a means to achieve these two goals. Genetic algorithm 疋-well-known random search algorithm, _ Yu in the huge solution to find a good solution. The basic step of the genetic algorithm is to evaluate the fitness of an individual from a group of wealth, and use this fitness as the evolution of mating and mutation between individuals. The evolution of each generation means a new child. One of the key points that generations might use the genetic algorithm is whether they can find the: r method. The so-called gene coding refers to how to turn the shots into turns (in the semiconductor crystal ® Gu Pai off shot is a set of rowing, and the scheduler of the side-side paste the second choice is slightly different-a batch If the shipment is a unit, if there is one for each batch of goods, the number of genes for the selection strategy is ^ integers, and each — indicates a machine selection strategy. The code for the priority of the shipment is — one machine = ,, Assume that there are “batch shipments” in the whole factory, and the number of general merchandise priority is equal to the number of integers and the number of excellent silks. — A complete gene = The machine selection strategy gene is concatenated with the gene of shipment preference for all machines = Cheng Bei The evaluation of Cheng is completed by the simulator of this patent. Due to the difficulty of simulating the preparation of fast wheels, it can be found in the exhaustion time. ^ _ Combination of Taiwan's selection strategy and the principle of the best area of approval In addition, the encoding method of this patent is compatible with the traditional health scheduling strategy, so

This paper size applies to China National Standard (CNS) A4 (210 X 297 ^ 17

丨 ί · _ installed -------- order --------- line · (Please read the note on the back? Matters before filling out this page) Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 583560 5. Description of the invention (10) This scheduler can still provide assistance in scheduling. Illustrated diagram Figure 1. Overall structure of the system This icon shows that the system is mainly divided into two parts, the simulator and the scheduler. Simulator The underlying model of the columnar Fischer network is built on top of the model. Figure 2 · The structure of the system's bottom model * This icon shows the three major models in the system and the sub-models included in each model. Figure 3. Traditional resource competition sub-network. Each batch must wait for the last batch to use the machine. After releasing the machine, it can enter the machine for processing. When traditional resource competition subnets face more than one waiting batch, they must often calculate the priority of the batch to select the next batch to be processed by the party. ~ Figure 4 · New-style resource queue subnet > When each batch ^ enters the queue of the machine, the waiting time for the batch of goods can be calculated immediately. After the waiting time passes, the batch of goods will naturally enter the machine. For processing. In the new poorly-sourced sub-network, the sub-network does not need to spend extra time in calculating the priority of goods and choosing the goods to be approved. In addition, the resource sub-network contains the -Nabe arrival rate update right group. The action of regularly updating the arrival rate reflects the flow of shipments in Wei, and also improves the accuracy of the waiting time formula value. The size of this paper is Ding Guanjiaxian (CNS) A4 (21G χ -JIU1I .------- · installation -------- order --------- line · (please First read the note on the back? Matters and then fill out this page) 583560 A7 B7 V. Description of the invention (11) Figure V. Schematic diagram of the change in the priority value of a batch of goods over time Assuming that the batch of goods enters the machine queue at time Γ, The priority level of the batch is corpse, and its corresponding priority parameter is ~ and ^, then the change of the priority value of the batch with time is shown in this figure. Figure 6. Flow chart of the batch work This figure is from the perspective of the batch Let's review the production process in the factory. When the batch is ready to enter the next process, the machine selection process must be performed first in order to select a machine in the machine group with the appropriate process capability and enter the machine. Waiting. If the machine is too far away from the current machine, you need to use the handling system. When the machine chooses to process the batch of goods, first check whether the pre-operation needs to be re-started. Second, in the process of processing the batch, it may also A machine stoppage occurred. At this time, the machine was repaired, and the batch of goods being processed was restarted. Select the appropriate machine outside the program. ★ Another figure VII. Batch selection flowchart This figure is a schematic diagram of the batch waiting on the machine selection list. The traditional system is to follow this process. However, in this patent office In the proposed system, these actions of calculating priority and selecting batches of goods have been calculated in the waiting time of the queue. Gang_Liqing Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs Employee Consumption Cooperative Figure VIII. Genetic algorithm Genes in the generographer's schematic diagram scheduler are mainly divided into two parts, which are priority levels. In the machine selection method, — the machine selection rules for batch shipments in a certain process, such as the shortest job priority method ^ 583560 A7 _B7_ V. Description of the invention (12), etc. In terms of priority level, each gene represents the priority level of a certain batch of goods on a certain machine. Figure IX. Flow chart of genetic algorithm This figure is used for Explain the operation flow of the genetic algorithm. The diagram is a simple illustration. The overall structure of the system is shown in Figure 2. The structure of the underlying model of the system is shown in Figure 3. The traditional sub-network of competitive resources is shown in Figure 4. Network diagram V. Schematic diagram of the change of priority value of batches over time. Figure 6. Flow chart of batch work. Figure 7. Flow chart of batch selection. Figure 8. Schematic diagram of genetic coding for genetic algorithm. Figure 9. Genetic algorithm Flow chart. 丨 ί: ------- • Installation -------- Order --------- Line # · (Please read the precautions on the back before filling this page) Economy Printed by the Ministry of Intellectual Property Bureau's Consumer Cooperatives, this paper is sized to the Chinese National Standard (CNS) A4 (210 X 297 mm)

Claims (1)

583560 6. Scope of patent application: 叫 called '丨 1 · A simulation / scheduling system applied to semiconductor wafer production. The system includes: A factory model, which is constructed using a Fischer network to describe Changes in the status of the batch of goods, machines and vehicles in the factory during the production process; Combined with Fischer's network and queuing theory, constructs a new resource competition subnet and arrival rate update submodule, and promotes a high-speed simulation system; A calculation formula that can directly calculate the waiting time for batches of each priority level in the factory; '' Equipped with a software system, after entering the relevant data required for manufacturing execution, it can output the batches and machines in the designated simulation time zone Various data of the station, such as production cycle, machine utilization and production process records, etc .; It has an optimization method that uses genetic algorithms as the solution search method. Under the specified schedule performance evaluation function, it can generate economic data. Optimize the priority level and machine selection method of the batch. 2. ^ Apply for the simulation / scheduling secret described in Item 丨 of the special fiber loop. The system's cheap model includes the machine working model and the batch handling model. 3. The simulation / scheduling system described in item 2 of the scope of patent application, the machine working model is characterized by including machine inspection, batch inspection and machine shutdown status. 4. The simulation / scheduling system as described in item 2 of Shenjing's patent scope, the characteristics of its batch handling model is to model the vehicle call, vehicle selection, vehicle transportation planning, vehicle routing, vehicle Route conflicts, vehicle movements, and more. 5. The simulation / scheduling system described in item i of Shenyan's patent scope. The system's software system provides functions such as factory operation simulation, work order delivery prediction, and wafer production manufacturing scheduling. 583560 9 · The simulation / scheduling system described in item 5 of the patent scope, in which the priority order of the batch processing of the batch λ! Of the machine is determined by the priority level of the batch and the waiting queue. Time is up. Ήw Inch ^ The simulation / scheduling system and system described in the 9th item of the sibling, where the parent-priority levels are mentioned, all have-for the parameters & and ~, each batch of shipments, Its priority is calculated by omission. ^ The simulation / scheduling system described in Mingshi Lishiweiang Item 0, which has Η Λ, which can calculate the waiting time for the shipment according to the parameters of the shipment priority level and the factory status. 12 · μ > 1. The simulation / scheduling system described in item 10 of Mingjin Liganwei, its priority calculation method is characterized by its ability to express in addition to the customizable priority calculation method. Sixth, the scope of patent application 丨 I__________________ — ---..... ~~ Several existing well-known priority rules, such as first-in, first-out, last-in, last-out, minimum remaining process priority, shortest delivery period priority, etc. 13. The simulation / scheduling system described in item 10 of the scope of patent application, wherein the set values of the parameters k and c can be used to reflect the scheduling strategy of the production manager. 14. The simulation / scheduling system described in item 1 of the scope of the patent application, in which the queuing theory is used to improve the original resource competition subnet. The batch of goods listed on each machine can be based on its priority level and The arrival speed of the batch of goods in the factory directly calculates the waiting time of the batch of goods in the queue, which is characterized by greatly reducing the time complexity of the simulation system operation. 15. The simulation / scheduling system described in item 13 of the scope of the patent application, where the plant model includes an arrival rate update sub-module, which will collect historical data and periodically update the arrival rate of shipments to improve the calculation of batches. Accuracy of cargo waiting time. 16. According to the simulation / scheduling system described in item 5 of the scope of patent application, the work order delivery time prediction function can start with the current in-plant status and predict the time when a new work order will be released. 17. The simulation / scheduling system described in item 5 of the scope of patent application, the wafer production and manufacturing scheduling function is achieved by a unique gene coding method, combined with genetic algorithm, and the system's own rapid simulation function. 18. The simulation / scheduling system described in item 16 of the scope of patent application, the gene coding method is composed of a resource gene and a work routing gene. 19. The resource gene described in item 17 of the scope of patent application, which indicates the priority level of a certain shipment on a certain machine. 20. The work routing gene described in item 17 of the scope of patent application, which represents a certain 583560 How to select the machine for each batch when selecting a certain process step. 21. The simulation / scheduling system described in item 16 of the scope of patent application, wherein the evaluation function of the genetic algorithm takes into account the total manufacturing time of the batch, the factory's productivity, and the delivery rate of the work order. 583560 Review opinion: The line in Figure 6 is not clear. Please correct it. Figure 6. Work flow chart for batch delivery